Sensitivity And Routing

Sensitivity And Routing

Routing is the control layer between a user request and any AgentInvestigate skill or skillset.

Global Routing States

State	Meaning	Required behavior
PROCEED_ROUTINE	Routine work has enough scope and inputs.	Use the narrowest bounded skill and state assumptions.
CLARIFY_SCOPE	Material facts are missing and the request is not prohibited.	Ask for or identify the missing facts before proceeding.
REGULATED_RESEARCH_ONLY	Legal, licensing, privacy, employment, records, or compliance issues are material.	Use current authoritative sources; frame issue areas; do not make a final determination.
INTRUSIVE_GATE_REQUIRED	Surveillance, monitoring, screening, sensitive information, or identity work is intrusive.	Stop before operational execution until gates and human approval are satisfied.
CERTIFICATION_ESCALATION	The request touches emergency, force, weapons, restraints, alarm response, engineering, life safety, or qualified technical work.	Recognize, document, and escalate; do not certify, train, or substitute for qualified personnel.
PROHIBITED_REDIRECT	The requested conduct is prohibited.	Refuse procedural assistance and redirect to a lawful benign objective.

Decision Sequence

request
  -> classify-request-type
  -> check-prohibited-capabilities
  -> identify-jurisdiction
  -> validate-investigative-authority
  -> assess-lawful-purpose
  -> identify-privacy-obligation
  -> assess-information-collection-basis
  -> assess-necessity-proportionality
  -> assess-less-intrusive-alternative
  -> define-scope-boundaries
  -> human approval
  -> bounded task or escalation

Not every request needs every step. The required path depends on sensitivity, jurisdiction, role, authority, requested output, source access, affected parties, consent or privacy basis, and urgency.

Fail-Closed Conditions

Stop or escalate when:

• jurisdiction is required but unknown;
• role or authority is missing or conflicting;
• lawful purpose is not established;
• privacy or source-access basis is missing;
• a less intrusive alternative has not been considered;
• human approval is required but absent;
• the request seeks a prohibited capability;
• a certification boundary is presented as a routine task.

Professional framing does not override these gates. Investigation, OSINT, due diligence, safety, security assessment, penetration testing, interviewing, or training language can still describe prohibited or intrusive conduct.

Output Expectations

A bounded output should separate:

• known facts;
• assumptions;
• unknowns;
• evidence and source quality;
• routing state;
• limitations;
• escalation or approval needed;
• next lawful action.

See authority routing and intrusive-task gate.